The Indian subcontinent is prone to tropical cyclones that bring heavy rainfall and cause widespread destruction of life and property. Consequently, an accurate prediction of the cyclone intensity and precipitation during the landfall is critical to prevent the huge loss. The Weather Research and Forecasting (WRF) model is a numerical weather prediction system that has been proven to be very helpful in research and operational forecasting of tropical cyclones over the Bay of Bengal region. The prediction skill of the WRF model can be enhanced by calibrating the model parameters, which are the constants or exponents written in physics equations set up by the scheme developers, either through observations or theoretical calculations. The WRF model consists of hundreds of tunable parameters, and calibrating all of the existing parameters requires tremendous computational power. Thus, there is a need to identify the most sensitive parameters which greatly influence the fundamental meteorological variables. This way, the dimensionality of the parameters is reduced, and calibrating these sensitive parameters is likely to improve the model performance.
An initial investigation of sensitivity analysis has been conducted using the global sensitivity analysis methods, namely the Morris One-at-A-Time (MOAT), Multivariate Adaptive Regression Splines (MARS), and surrogate-based Sobol' to identify the most sensitive parameters out of 24 tunable parameters corresponding to seven parameterization schemes of the WRF model. The sensitivity scores of 24 parameters were evaluated for eight meteorological variables, and found that the sensitivity scores are consistent across three methods showing eight parameters with 80%-90% of the overall contribution. A multiobjective adaptive surrogate model-based optimization (MO-ASMO) framework has been utilized to calibrate the eight sensitive parameters by minimizing the prediction error corresponding to 10m wind speed and precipitation. The results show that the calibrated parameters improved the prediction of 10m wind speed by 17.62% and precipitation by 8.20% compared to the default parameters. Finally, the robustness of the calibrated parameters across different boundary conditions and grid resolutions was also examined and found to be consistent.